Re: [PATCH 3/6] x86: Add nopv parameter to disable PV extensions

From: Zhenzhong Duan
Date: Mon Jun 24 2019 - 09:50:56 EST

On 2019/6/24 21:18, Juergen Gross wrote:
On 23.06.19 15:01, Zhenzhong Duan wrote:
In virtualization environment, PV extensions (drivers, interrupts,
timers, etc) are enabled in the majority of use cases which is the
best option.

However, in some cases (kexec not fully working, benchmarking)
we want to disable PV extensions. As such introduce the
'nopv' parameter that will do it.

There is already 'xen_nopv' parameter for XEN platform but not for
others. 'xen_nopv' can then be removed with this change.

Signed-off-by: Zhenzhong Duan <zhenzhong.duan@xxxxxxxxxx>
Cc: xen-devel@xxxxxxxxxxxxxxxxxxxx
 Documentation/admin-guide/kernel-parameters.txt | 4 ++++
 arch/x86/kernel/cpu/hypervisor.c | 11 +++++++++++
 2 files changed, 15 insertions(+)

diff --git a/Documentation/admin-guide/kernel-parameters.txt b/Documentation/admin-guide/kernel-parameters.txt
index 138f666..b352f36 100644
--- a/Documentation/admin-guide/kernel-parameters.txt
+++ b/Documentation/admin-guide/kernel-parameters.txt
@@ -5268,6 +5268,10 @@
ÂÂÂÂÂÂÂÂÂÂÂÂÂ improve timer resolution at the expense of processing
ÂÂÂÂÂÂÂÂÂÂÂÂÂ more timer interrupts.
 + nopv= [X86]
+ÂÂÂÂÂÂÂÂÂÂÂ Disables the PV optimizations forcing the guest to run
+ÂÂÂÂÂÂÂÂÂÂÂ as generic guest with no PV drivers.
diff --git a/arch/x86/kernel/cpu/hypervisor.c b/arch/x86/kernel/cpu/hypervisor.c
index 479ca47..4f2c875 100644
--- a/arch/x86/kernel/cpu/hypervisor.c
+++ b/arch/x86/kernel/cpu/hypervisor.c
@@ -85,10 +85,21 @@ static void __init copy_array(const void *src, void *target, unsigned int size)
ÂÂÂÂÂÂÂÂÂÂÂÂÂ to[i] = from[i];
 +static bool nopv;
+static __init int xen_parse_nopv(char *arg)
+ÂÂÂ nopv = true;
+ÂÂÂ return 0;
+early_param("nopv", xen_parse_nopv);
 void __init init_hypervisor_platform(void)
ÂÂÂÂÂ const struct hypervisor_x86 *h;
 + if (nopv)
+ÂÂÂÂÂÂÂ return;

Oh, this is no good idea.

There are guest types which just won't work without pv interfaces, like
Xen PV and Xen PVH. Letting them fail due to just a wrong command line
parameter is not nice, especially as the failure might be very hard to
track down to the issue for the user.
Yes, thanks for catching.

I guess you could add a "ignore_nopv" member to struct hypervisor_x86
set to true for the mentioned guest types and call the detect functions
only if nopv is false or ignore_nopv is true.

I think your suggestion is good, I'll rework it based on your suggestion.